The increasing demand for autonomous space operations motivates lightweight AI-based approaches to Resident Space Objects (RSO) inspection, where spacecraft must balance trajectory optimization with strict onboard resource limitations in real-time. This paper presents a Deep Q-Network (DQN) for single-agent RSO inspection, combining fuel-efficient orbital exploration with battery management and data down-linking in a unified control policy. The environment features a discrete set of relative orbits, stochastic resource dynamics, and a closed-loop transfer method based on Relative Orbital Elements (ROEs). Results show how, across multiple training runs, the agent learns to accomplish the mission with a low number of transfers while efficiently handling battery and data processes.
Single Agent On-Orbit Inspection: Energy-Awareness, Data Down-linking and Orbital Exploration via Deep-Q-Learning
Francesco Paolo SALZO;Giordana BUCCHIONI
2025-01-01
Abstract
The increasing demand for autonomous space operations motivates lightweight AI-based approaches to Resident Space Objects (RSO) inspection, where spacecraft must balance trajectory optimization with strict onboard resource limitations in real-time. This paper presents a Deep Q-Network (DQN) for single-agent RSO inspection, combining fuel-efficient orbital exploration with battery management and data down-linking in a unified control policy. The environment features a discrete set of relative orbits, stochastic resource dynamics, and a closed-loop transfer method based on Relative Orbital Elements (ROEs). Results show how, across multiple training runs, the agent learns to accomplish the mission with a low number of transfers while efficiently handling battery and data processes.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
